Data Preprocessing

library(tidyverse) ; library(reshape2) ; library(glue) ; library(plotly) ; library(plotlyutils)
library(RColorBrewer) ; library(viridis) ; require(gridExtra) ; library(GGally) ; library(ggExtra) ; library(ggpubr)
library(biomaRt) ; library(DESeq2) ; library(sva) ; library(WGCNA) ; library(vsn)
library(dendextend) ; library(expss)
library(knitr) ; library(kableExtra)

---

Raw data

Dataset downloaded from mgandal’s github repository.

Load and annotate data

# Load csvs
datExpr = read.csv('./../Data/RNAseq_ASD_datExpr.csv', row.names=1)
datMeta = read.csv('./../Data/RNAseq_ASD_datMeta.csv')

# 1. Group brain regions by lobes
# 2. Remove '/' from Batch variable: (It is recommended (but not required) to use only letters, numbers, 
#    and delimiters '_' or '.', in levels of factors as these are safe characters for column names in R
# 3. Transform Diagnosis into a factor variable
datMeta = datMeta %>% mutate(Brain_Region = as.factor(Region)) %>% 
                      mutate(Brain_lobe = ifelse(Brain_Region %in% c('BA4_6', 'BA9', 'BA24', 'BA44_45'), 
                                                 'Frontal',
                                          ifelse(Brain_Region %in% c('BA3_1_2_5', 'BA7'), 'Parietal',
                                          ifelse(Brain_Region %in% c('BA38','BA39_40','BA20_37','BA41_42_22'), 
                                                 'Temporal',
                                          'Occipital')))) %>%
                      mutate(Batch = as.factor(gsub('/', '.', RNAExtractionBatch)), 
                             Diagnosis = factor(Diagnosis_, levels=c('CTL','ASD'))) %>% 
                      dplyr::select(-Diagnosis_)


# GO Neuronal annotations: regex 'neuron' in GO functional annotations and label the genes that make a match as neuronal
GO_annotations = read.csv('./../Data/genes_GO_annotations.csv')
GO_neuronal = GO_annotations %>% filter(grepl('neuron', go_term)) %>% 
              mutate('ID'=as.character(ensembl_gene_id)) %>% 
              dplyr::select(-ensembl_gene_id) %>% distinct(ID) %>%
              mutate('Neuronal'=1)


# SFARI Genes
SFARI_genes = read_csv('./../../../SFARI/Data/SFARI_genes_01-03-2020_w_ensembl_IDs.csv')
SFARI_genes = SFARI_genes[!duplicated(SFARI_genes$ID) & !is.na(SFARI_genes$ID),]

# NCBI biotype annotation
NCBI_biotype = read.csv('./../../../NCBI/Data/gene_biotype_info.csv') %>% 
               dplyr::rename('ensembl_gene_id'=Ensembl_gene_identifier, 'gene_biotype'=type_of_gene, 
                             'hgnc_symbol'=Symbol) %>% 
               mutate(gene_biotype = ifelse(gene_biotype=='protein-coding','protein_coding', gene_biotype))


rm(GO_annotations)

Check sample composition

Data description taken from the dataset’s synapse entry: RNAseq data was generated from 88 postmortem cortex brain samples from subjects with ASD (53 samples from 24 subjects) and non-psychiatric controls (35 samples from 17 subjects), across four cortical regions encompassing all major cortical lobes – frontal, temporal, parietal, and occipital. Brain samples were obtained from the Harvard Brain Bank as part of the Autism Tissue Project (ATP).

The dataset includes 63682 genes from 88 samples belonging to 41 different subjects

Counts distribution: More than half of the counts are zero and most of the counts are relatively low, but there are some very high outliers

counts = datExpr %>% melt

count_distr = data.frame('Statistic' = c('Min', '1st Quartile', 'Median', 'Mean', '3rd Quartile', 'Max'),
                         'Values' = c(min(counts$value), quantile(counts$value, probs = c(.25, .5)) %>% unname,
                                      mean(counts$value), quantile(counts$value, probs = c(.75)) %>% unname,
                                      max(counts$value)))

count_distr %>% kable(digits = 2, format.args = list(scientific = FALSE)) %>% kable_styling(full_width = F)

Statistic	Values
Min	0.00
1st Quartile	0.00
Median	0.00
Mean	564.09
3rd Quartile	27.00
Max	27183314.00

rm(counts, count_distr)

Diagnosis distribution by Sample: There are more ASD samples than controls

table_info = datMeta %>% apply_labels(Diagnosis = 'Diagnosis', Brain_lobe = 'Brain Lobe', Batch = 'Batch', Sex = 'Gender')
cro(table_info$Diagnosis)

	#Total
Diagnosis
CTL	35
ASD	53
#Total cases	88

Diagnosis distribution by Subject: There are more ASD patients than controls

cro(table_info$Diagnosis[!duplicated(table_info$Subject_ID)])

	#Total
Diagnosis
CTL	17
ASD	24
#Total cases	41

Brain region distribution: All regions are balanced

cro(table_info$Brain_lobe)

	#Total
Brain Lobe
Frontal	22
Occipital	23
Parietal	23
Temporal	20
#Total cases	88

Diagnosis and brain region seem to be balanced except for the frontal lobe, where there are more control samples than ASD ones

cro(table_info$Diagnosis, list(table_info$Brain_lobe,total()))

	Brain Lobe					#Total
	Frontal	Occipital	Parietal	Temporal
Diagnosis
CTL	13	8	8	6		35
ASD	9	15	15	14		53
#Total cases	22	23	23	20		88

Sex distribution: There are many more Male samples than Female ones

cro(table_info$Sex)

	#Total
Gender
F	15
M	73
#Total cases	88

Diagnosis and Gender seem to be balanced

cro(table_info$Diagnosis, list(table_info$Sex, total()))

	Gender			#Total
	F	M
Diagnosis
CTL	6	29		35
ASD	9	44		53
#Total cases	15	73		88

Age distribution: Subjects between 2 and 60 years old with a mean close to 30

summary(datMeta$Age)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##    2.00   17.00   28.00   29.74   41.75   60.00

Annotate genes with BioMart information

I was originally running this with the feb2014 version of BioMart because that’s the one that Gandal used (and it finds all of the Ensembl IDs, which other versions don’t), but it has some outdated biotype annotations, to fix this I’ll obtain all the information except the biotype label from BioMart in the same way as it had been done before, and then I’ll add the most current biotype label using information from NCBI’s website and then from BioMart in the following way:

1. Use BioMart to run a query with the original feb2014 version using the Ensembl IDs as keys to obtain all the information except the biotype labels

2. Annotate genes with Biotype labels:

   2.1 Use the NCBI annotations downloaded from NCBI’s website and processed in NCBI/RMarkdowns/clean_data.html (there is information for only 26K genes, so some genes will remain unlabelled)

   2.2 Use the current version (jan2020) to obtain the biotype annotations using the Ensembl ID as keys (some genes don’t return a match)

   2.3 For the genes that didn’t return a match, use the current version (jan2020) to obtain the biotype annotations using the gene name as keys (17 genes return multiple labels)

   2.4 For the genes that returned multiple labels, use the feb2014 version with the Ensembl IDs as keys

labels_source = data.frame('source' = c('NCBI', 'BioMart2020_byID', 'BioMart2020_byGene', 'BioMart2014'),
                                      'n_matches' = rep(0,4))

########################################################################################
# 1. Query archive version

getinfo = c('ensembl_gene_id','external_gene_id','chromosome_name','start_position',
            'end_position','strand')
mart = useMart(biomart = 'ENSEMBL_MART_ENSEMBL', dataset = 'hsapiens_gene_ensembl', 
               host = 'feb2014.archive.ensembl.org')
datGenes = getBM(attributes = getinfo, filters=c('ensembl_gene_id'), values = rownames(datExpr), mart=mart) %>% 
           rename(external_gene_id = 'hgnc_symbol')
datGenes$length = datGenes$end_position - datGenes$start_position

cat(paste0('1. ', sum(is.na(datGenes$start_position)), '/', nrow(datGenes),
             ' Ensembl IDs weren\'t found in the feb2014 version of BioMart'))

## 1. 0/63677 Ensembl IDs weren't found in the feb2014 version of BioMart

########################################################################################
########################################################################################
# 2. Get Biotype Labels

cat('2. Add biotype information')

## 2. Add biotype information

########################################################################################
# 2.1 Add NCBI annotations
datGenes = datGenes %>% left_join(NCBI_biotype, by=c('ensembl_gene_id','hgnc_symbol'))

cat(paste0('2.1 ' , sum(is.na(datGenes$gene_biotype)), '/', nrow(datGenes),
             ' Ensembl IDs weren\'t found in the NCBI database'))

## 2.1 42904/63677 Ensembl IDs weren't found in the NCBI database

labels_source$n_matches[1] = sum(!is.na(datGenes$gene_biotype))

########################################################################################
# 2.2 Query current BioMart version for gene_biotype using Ensembl ID as key

getinfo = c('ensembl_gene_id','gene_biotype')
mart = useMart(biomart = 'ENSEMBL_MART_ENSEMBL', dataset = 'hsapiens_gene_ensembl',
               host = 'jan2020.archive.ensembl.org')
datGenes_biotype = getBM(attributes = getinfo, filters = c('ensembl_gene_id'), mart=mart, 
                         values = datGenes$ensembl_gene_id[is.na(datGenes$gene_biotype)])

cat(paste0('2.2 ' , sum(is.na(datGenes$gene_biotype))-nrow(datGenes_biotype), '/', 
           sum(is.na(datGenes$gene_biotype)),
           ' Ensembl IDs weren\'t found in the jan2020 version of BioMart when querying by Ensembl ID'))

## 2.2 9099/42904 Ensembl IDs weren't found in the jan2020 version of BioMart when querying by Ensembl ID

# Add new gene_biotype info to datGenes
datGenes = datGenes %>% left_join(datGenes_biotype, by='ensembl_gene_id') %>%
           mutate(gene_biotype = coalesce(as.character(gene_biotype.x), gene_biotype.y)) %>%
           dplyr::select(-gene_biotype.x, -gene_biotype.y)

labels_source$n_matches[2] = sum(!is.na(datGenes$gene_biotype)) - labels_source$n_matches[1]

########################################################################################
# 3. Query current BioMart version for gene_biotype using gene symbol as key

missing_genes = unique(datGenes$hgnc_symbol[is.na(datGenes$gene_biotype)])
getinfo = c('hgnc_symbol','gene_biotype')
datGenes_biotype_by_gene = getBM(attributes=getinfo, filters=c('hgnc_symbol'), mart=mart,
                                 values=missing_genes)

cat(paste0('2.3 ', length(missing_genes)-length(unique(datGenes_biotype_by_gene$hgnc_symbol)),'/',
           length(missing_genes),
           ' genes weren\'t found in the current BioMart version when querying by gene name'))

## 2.3 5712/7866 genes weren't found in the current BioMart version when querying by gene name

dups = unique(datGenes_biotype_by_gene$hgnc_symbol[duplicated(datGenes_biotype_by_gene$hgnc_symbol)])
cat(paste0('    ', length(dups), ' genes returned multiple labels (these won\'t be added)'))

##     17 genes returned multiple labels (these won't be added)

# Update information
datGenes_biotype_by_gene = datGenes_biotype_by_gene %>% filter(!hgnc_symbol %in% dups)
datGenes = datGenes %>% left_join(datGenes_biotype_by_gene, by='hgnc_symbol') %>% 
            mutate(gene_biotype = coalesce(gene_biotype.x, gene_biotype.y)) %>%
            dplyr::select(-gene_biotype.x, -gene_biotype.y)

labels_source$n_matches[3] = sum(!is.na(datGenes$gene_biotype)) - sum(labels_source$n_matches)

########################################################################################
# 4. Query feb2014 BioMart version for the missing biotypes

missing_ensembl_ids = unique(datGenes$ensembl_gene_id[is.na(datGenes$gene_biotype)])

getinfo = c('ensembl_gene_id','gene_biotype')
mart = useMart(biomart = 'ENSEMBL_MART_ENSEMBL', dataset = 'hsapiens_gene_ensembl', 
               host = 'feb2014.archive.ensembl.org')
datGenes_biotype_archive = getBM(attributes = getinfo, filters=c('ensembl_gene_id'), 
                                 values = missing_ensembl_ids, mart=mart)

cat(paste0('2.4 ', length(missing_ensembl_ids)-nrow(datGenes_biotype_archive),'/',length(missing_ensembl_ids),
             ' genes weren\'t found in the feb2014 BioMart version when querying by Ensembl ID'))

## 2.4 0/6648 genes weren't found in the feb2014 BioMart version when querying by Ensembl ID

# Update information
datGenes = datGenes %>% left_join(datGenes_biotype_archive, by='ensembl_gene_id') %>% 
            mutate(gene_biotype = coalesce(gene_biotype.x, gene_biotype.y)) %>%
            dplyr::select(-gene_biotype.x, -gene_biotype.y)

labels_source$n_matches[4] = sum(!is.na(datGenes$gene_biotype)) - sum(labels_source$n_matches)

########################################################################################
# Plot results

labels_source = labels_source %>% mutate(x = 1, percentage = round(100*n_matches/sum(n_matches),1))

p = labels_source %>% ggplot(aes(x, percentage, fill=source)) + geom_bar(position='stack', stat='identity') +
    theme_minimal() + coord_flip() + theme(legend.position='bottom', axis.title.y=element_blank(),
    axis.text.y=element_blank(), axis.ticks.y=element_blank())

ggplotly(p + theme(legend.position='none'))

as_ggplot(get_legend(p))

########################################################################################
# Reorder rows to match datExpr
datGenes = datGenes[match(rownames(datExpr), datGenes$ensembl_gene_id),]


rm(getinfo, mart, datGenes_biotype, datGenes_biotype_by_gene, datGenes_biotype_archive,
   dups, missing_ensembl_ids, missing_genes, labels_source, p)

---

Filtering

Checking how many SFARI genes are in the dataset

df = SFARI_genes %>% dplyr::select(-gene_biotype) %>% inner_join(datGenes, by=c('ID'='ensembl_gene_id'))
n_SFARI = df[['gene-symbol']] %>% unique %>% length

Considering all genes, this dataset contains 911 of the 912 SFARI genes

1.- Filter entries that don’t correspond to genes

to_keep = !is.na(datGenes$length)

Names of the rows removed: __no_feature, __ambiguous, __too_low_aQual, __not_aligned, __alignment_not_unique

datGenes = datGenes[to_keep,]
datExpr = datExpr[to_keep,]
rownames(datGenes) = datGenes$ensembl_gene_id

Removed 5 ‘genes’, 63677 remaining

2. Filter genes that do not encode any protein

35% of the genes are protein coding genes

datGenes$gene_biotype %>% table %>% sort(decreasing=TRUE) %>% kable(caption='Biotypes of genes in dataset') %>%
                          kable_styling(full_width = F)

Biotypes of genes in dataset

.	Freq
protein_coding	22543
lncRNA	12167
processed_pseudogene	10117
unprocessed_pseudogene	2547
1	2314
miRNA	2276
misc_RNA	2178
snRNA	2043
pseudogene	1410
snoRNA	1202
lincRNA	840
transcribed_unprocessed_pseudogene	682
rRNA_pseudogene	500
transcribed_processed_pseudogene	441
antisense	380
3	331
6	314
IG_V_pseudogene	254
IG_V_gene	179
TR_V_gene	146
transcribed_unitary_pseudogene	86
TR_J_gene	81
unitary_pseudogene	74
processed_transcript	72
sense_intronic	72
IG_D_gene	64
rRNA	49
TR_V_pseudogene	46
sense_overlapping	38
scaRNA	31
polymorphic_pseudogene	28
7	25
IG_J_gene	24
IG_C_gene	23
Mt_tRNA	22
4	17
IG_C_pseudogene	11
TEC	11
TR_C_gene	8
3prime_overlapping_ncrna	6
IG_J_pseudogene	6
ribozyme	5
TR_J_pseudogene	5
TR_D_gene	3
Mt_rRNA	2
8	1
translated_processed_pseudogene	1
translated_unprocessed_pseudogene	1
vaultRNA	1

Most of the non-protein coding genes have very low levels of expression

plot_data = data.frame('ID' = rownames(datExpr), 'MeanExpr' = apply(datExpr, 1, mean),
                       'ProteinCoding' = datGenes$gene_biotype=='protein_coding')

ggplotly(plot_data %>% ggplot(aes(log2(MeanExpr+1), fill=ProteinCoding, color=ProteinCoding)) + 
         geom_density(alpha=0.5) + theme_minimal())

rm(plot_data)

df = SFARI_genes %>% dplyr::select(-gene_biotype) %>% inner_join(datGenes, by=c('ID'='ensembl_gene_id'))

Filtering protein coding genes, we are left with 907 SFARI Genes (we lose 4 genes)

Note: The gene name for Ensembl ID ENSG00000187951 is wrong, it should be AC091057.1 instead of ARHGAP11B, but the biotype is right, so it would still be filtered out

n_SFARI = df[['gene-symbol']][df$gene_biotype=='protein_coding'] %>% unique %>% length

df %>% filter(!`gene-symbol` %in% df$`gene-symbol`[df$gene_biotype=='protein_coding']) %>% 
       dplyr::select(ID, `gene-symbol`, `gene-score`, gene_biotype, syndromic, `number-of-reports`) %>% 
       kable(caption='Lost Genes')  %>% kable_styling(full_width = F)

Lost Genes

ID	gene-symbol	gene-score	gene_biotype	syndromic	number-of-reports
ENSG00000187951	ARHGAP11B	3	lncRNA	0	2
ENSG00000251593	MSNP1AS	2	processed_pseudogene	0	12
ENSG00000233067	PTCHD1-AS	2	lncRNA	0	3
ENSG00000197558	SSPO	3	transcribed_unitary_pseudogene	0	4

rm(df)

if(!all(rownames(datExpr)==rownames(datGenes))) cat('!!! gene rownames do not match!!!')

to_keep = datGenes$gene_biotype=='protein_coding'
datExpr = datExpr %>% filter(to_keep)
datGenes = datGenes %>% filter(to_keep)
rownames(datExpr) = datGenes$ensembl_gene_id
rownames(datGenes) = datGenes$ensembl_gene_id

Removed 41134 genes. 22543 remaining

3. Filter genes with low expression levels

\(\qquad\) 3.1 Remove genes with zero expression in all of the samples

to_keep = rowSums(datExpr) > 0

df = data.frame('rowSums' = rowSums(datExpr), 'ensembl_gene_id' = rownames(datExpr)) %>%
     right_join(SFARI_genes, by='ensembl_gene_id') %>% filter(rowSums == 0 & !is.na(`gene-score`)) %>%
     arrange(`gene-score`) %>% dplyr::select(-ensembl_gene_id) %>%
     filter(!duplicated(`gene-symbol`), !`gene-symbol` %in% datGenes$hgnc_symbol[to_keep])

datExpr = datExpr[to_keep,]
datGenes = datGenes[to_keep,]

Removed 3368 genes. 19175 remaining

904 SFARI genes remaining (we lost 3 genes)

n_SFARI = SFARI_genes[['gene-symbol']][SFARI_genes$ID %in% rownames(datExpr)] %>% unique %>% length

df %>% dplyr::select(ID, `gene-symbol`, `gene-score`, gene_biotype, syndromic, `number-of-reports`) %>% 
       kable(caption='Lost Genes with Top Scores')  %>% kable_styling(full_width = F)

Lost Genes with Top Scores

ID	gene-symbol	gene-score	gene_biotype	syndromic	number-of-reports
ENSG00000235718	MFRP	2	protein_coding	0	6
ENSG00000186393	KRT26	3	protein_coding	0	2
ENSG00000221888	OR1C1	3	protein_coding	0	3

rm(df)

\(\qquad\) 3.2 Removing genes with a high percentage of zeros

Choosing the threshold:

Criteria for selecting the percentage of zeros threshold: The minimum value in which the preprocessed data is relatively homoscedastic (we’re trying to get rid of the group of genes with very low mean and SD that make the cloud of points look like a comic book speech bubble)

• On the plot I’m using the “dual” of the maximum percentage of zeros, the minimum percentage of non zeros so the visualisation is more intuitive

• 75% seems to be a good threshold for the minimum percentage of non zeros, so 25% will be the maximum percentage of zeros allowed in a row

• The Mean vs SD plot doesn’t show all of the genes, a random sample was selected for the genes with higher level of expression so the visualisation wouldn’t be as heavy (and since we care about the genes with the lowest levels of expression, we aren’t losing important information)

datMeta_original = datMeta
datExpr_original = datExpr
datGenes_original = datGenes

# Return to original variables
datExpr = datExpr_original
datGenes = datGenes_original
datMeta = datMeta_original

rm(datExpr_original, datGenes_original, datMeta_original, datExpr_vst, datGenes_vst, datMeta_vst)

Selecting a threshold of 75

# Minimum percentage of non-zero entries allowed per gene
threshold = 75

plot_data = data.frame('id'=rownames(datExpr),
                       'non_zero_percentage' = apply(datExpr, 1, function(x) 100*mean(x>0)))

ggplotly(plot_data %>% ggplot(aes(x=non_zero_percentage)) + 
         geom_density(color='#0099cc', fill='#0099cc', alpha=0.3) + 
         geom_vline(xintercept=threshold, color='gray') +
         ggtitle('Percentage of non-zero entries distribution') + theme_minimal())

to_keep = apply(datExpr, 1, function(x) 100*mean(x>0)) >= threshold
datGenes = datGenes[to_keep,]
datExpr = datExpr[to_keep,]

Removed 3028 genes. 16147 remaining

864 SFARI genes remaining (we lost 40 genes)

n_SFARI = SFARI_genes[['gene-symbol']][SFARI_genes$ID %in% rownames(datExpr)] %>% unique %>% length

rm(threshold, plot_data, to_keep)

4. Filter outlier samples

\(\qquad\) 4.1 Gandal filters samples belonging to subject AN03345 without giving an explanation. Since it could have some technical problems, I will remove them as well

to_keep = (datMeta$Subject_ID != 'AN03345')
datMeta = datMeta[to_keep,]
datExpr = datExpr[,to_keep]

Removed 2 samples 86 remaining

\(\qquad\) 4.2 Filter out outliers: Using node connectivity as a distance measure, normalising it and filtering out genes farther away than 2 standard deviations from the left (lower connectivity than average, not higher)

• Gandal uses the formula \(s_{ij}=\frac{1+bw(i,j)}{2}\) to convert all the weights to positive values, but I used \(s_{ij}=|bw(i,j)|\) instead because I think it makes more sense. In the end it doesn’t matter because they select as outliers the same six samples

• All the outlier samples belong to the ASD group. Apart from this they don’t seem to have any characterstic in common (different subjects, extraction batches, brain lobes, age, PMI), except for Sex (but this is probably just because of the sex bias in the dataset)

absadj = datExpr %>% bicor %>% abs
netsummary = fundamentalNetworkConcepts(absadj)
ku = netsummary$Connectivity
z.ku = (ku-mean(ku))/sqrt(var(ku))

plot_data = data.frame('sample'=1:length(z.ku), 'distance'=z.ku, 'Sample_ID'=datMeta$Sample_ID, 
                       'Subject_ID'=datMeta$Subject_ID, 'Extraction_Batch'=datMeta$RNAExtractionBatch,
                       'Brain_Lobe'=datMeta$Brain_lobe, 'Sex'=datMeta$Sex, 'Age'=datMeta$Age,
                       'Diagnosis'=datMeta$Diagnosis, 'PMI'=datMeta$PMI)

selectable_scatter_plot(plot_data, plot_data[,-c(1:3)])

Outlier samples: AN01971_BA38, AN17254_BA17, AN09714_BA38, AN01093_BA7, AN02987_BA17, AN11796_BA7

to_keep = z.ku > -2
datMeta = datMeta[to_keep,]
datExpr = datExpr[,to_keep]

rm(absadj, netsummary, ku, z.ku, plot_data)

Removed 6 samples, 80 remaining

rm(to_keep)

5. Filter repeated genes

There are 15 genes with more than one ensembl ID in the dataset. To accurately refer to the rows of my data as ‘genes’, I’m going to remove the repeated ones.

dup_genes = datGenes$hgnc_symbol %>% duplicated

datGenes = datGenes[!dup_genes,]
datExpr = datExpr[!dup_genes,]

Removed 15 genes. 16132 remaining

864 SFARI genes remaining (we lost 0 genes)

rm(dup_genes, n_SFARI)

After filtering, the dataset consists of 16132 genes and 80 samples

Save filtered and annotated dataset

save(datExpr, datMeta, datGenes, file='./../Data/filtered_raw_data.RData')
#load('./../Data/filtered_raw_data.RData')

---

Batch Effects Exploratory Analysis

Note: No batch correction is performed in this section, this is done after the normalisation step

According to Tackling the widespread and critical impact of batch effects in high-throughput data, technical artifacts can be an important source of variability in the data, so batch correction should be part of the standard preprocessing pipeline of gene expression data.

They say Processing group and Date of the experiment are good batch surrogates, so I’m going to see if they affect the data in any clear way to use them as surrogates.

Processing group

All the information we have is the Brain Bank, and although all the samples were obtained from the Autism Tissue Project, we don’t have any more specific information about who preprocessed each sample

table_info = datMeta %>% apply_labels(Brain_Bank = 'Brain Bank', RNAExtractionBatch = 'RNA Extraction Batch',
                                      Diagnosis = 'Diagnosis')

table_info$Brain_Bank %>% cro

	#Total
Brain Bank
ATP	80
#Total cases	80

Date of processing

There are two different dates when the data was procesed

table_info$RNAExtractionBatch %>% cro

	#Total
RNA Extraction Batch
10/10/2014	53
6/20/2014	27
#Total cases	80

Luckily, there doesn’t seem to be a correlation between the batch surrogate and the objective variable, so the batch effect will not get confused with the Diagnosis effect

cro(table_info$RNAExtractionBatch, table_info$Diagnosis)

	Diagnosis
	CTL	ASD
RNA Extraction Batch
10/10/2014	24	29
6/20/2014	11	16
#Total cases	35	45

rm(table_info)

*All the samples from each subject were processed on the same day

Samples don’t seem to cluster together that strongly by batch, although there does seem to be some kind of relation

h_clusts = datExpr %>% t %>% dist %>% hclust %>% as.dendrogram

create_viridis_dict = function(){
  min_age = datMeta$Age %>% min
  max_age = datMeta$Age %>% max
  viridis_age_cols = viridis(max_age - min_age + 1)
  names(viridis_age_cols) = seq(min_age, max_age)
  
  return(viridis_age_cols)
}
viridis_age_cols = create_viridis_dict()

dend_meta = datMeta[match(substring(labels(h_clusts),2), datMeta$Dissected_Sample_ID),] %>% 
            mutate('Batch' = ifelse(RNAExtractionBatch=='10/10/2014', '#F8766D', '#00BFC4'),
                   'Diagnosis' = ifelse(Diagnosis=='CTL','#008080','#86b300'), # Blue control, Green ASD
                   'Sex' = ifelse(Sex=='F','#ff6666','#008ae6'),                # Pink Female, Blue Male
                   'Region' = case_when(Brain_lobe=='Frontal'~'#F8766D',        # ggplot defaults for 4 colours
                                        Brain_lobe=='Temporal'~'#7CAE00',
                                        Brain_lobe=='Parietal'~'#00BFC4',
                                        Brain_lobe=='Occipital'~'#C77CFF'),
                   'Age' = viridis_age_cols[as.character(Age)]) %>%             # Purple: young, Yellow: old
            dplyr::select(Age, Region, Sex, Diagnosis, Batch)
h_clusts %>% as.dendrogram %>% dendextend::set('labels', rep('', nrow(datMeta))) %>% 
             dendextend::set('branches_k_color', k=9) %>% plot
colored_bars(colors=dend_meta)

rm(h_clusts, dend_meta, create_viridis_dict, viridis_age_cols)

Comparing the mean expression of each sample by batch we can see there could be some batch effect differentiating them

plot_data_b1 = data.frame('Mean'=colMeans(datExpr[,datMeta$RNAExtractionBatch=='10/10/2014']), 
                          'Batch'='10/10/2014')
plot_data_b2 = data.frame('Mean'=colMeans(datExpr[,datMeta$RNAExtractionBatch=='6/20/2014']), 
                          'Batch'='6/20/2014')

plot_data = rbind(plot_data_b1, plot_data_b2)
mu = plot_data %>% group_by(Batch) %>% dplyr::summarise(BatchMean=mean(Mean))

ggplotly(plot_data %>% ggplot(aes(x=Mean, color=Batch, fill=Batch)) + geom_density(alpha=0.3) + 
         geom_vline(data=mu, aes(xintercept=BatchMean, color=Batch), linetype='dashed') +
         ggtitle('Mean expression by sample grouped by Batch') + scale_x_log10() + theme_minimal())

rm(plot_data_b1, plot_data_b2, plot_data, mu)

Looking for unknown sources of batch effects

Following the pipeline from Surrogate variable analysis: hidden batch effects where sva is used with DESeq2.

Create a DeseqDataSet object, estimate the library size correction and save the normalized counts matrix

counts = datExpr %>% as.matrix
rowRanges = GRanges(datGenes$chromosome_name,
                  IRanges(datGenes$start_position, width=datGenes$length),
                  strand=datGenes$strand,
                  feature_id=datGenes$ensembl_gene_id)
se = SummarizedExperiment(assays=SimpleList(counts=counts), rowRanges=rowRanges, colData=datMeta)
dds = DESeqDataSet(se, design = ~ Diagnosis)

dds = estimateSizeFactors(dds)
norm.cts = counts(dds, normalized = TRUE)

Provide the normalized counts and two model matrices to SVA. The first matrix uses the biological condition, and the second model matrix is the null model.

mod = model.matrix(~ Diagnosis, colData(dds))
mod0 = model.matrix(~ 1, colData(dds))
sva_fit = svaseq(norm.cts, mod=mod, mod0=mod0)

## Number of significant surrogate variables is:  13 
## Iteration (out of 5 ):1  2  3  4  5

rm(mod, mod0, norm.cts)

Found 13 surrogate variables, since there is no direct way to select which ones to pick Bioconductor answer, kept all of them.

Include SV estimations to datMeta information

sv_data = sva_fit$sv %>% data.frame
colnames(sv_data) = paste0('SV', 1:ncol(sv_data))

datMeta_sva = cbind(datMeta, sv_data)

rm(sv_data, sva_fit)

In conclusion: Date of extraction works as a surrogate for batch effect and the sva package found other 13 variables that could work as surrogates which are now included in datMeta and should be included in the DEA.

---

Normalisation and Differential Expression Analysis

Using DESeq2 package to perform normalisation. I chose this package over limma because limma uses the log transformed data as input instead of the raw counts and I have discovered that in this dataset, this transformation affects genes differently depending on their mean expression level, and genes with a high SFARI score are specially affected by this.

plot_data = data.frame('ID'=rownames(datExpr), 'Mean'=rowMeans(datExpr), 'SD'=apply(datExpr,1,sd))

plot_data %>% ggplot(aes(Mean, SD)) + geom_point(color='#0099cc', alpha=0.1) + geom_abline(color='gray') +
              scale_x_log10() + scale_y_log10() + theme_minimal()

rm(plot_data)

• Using vst instead of rlog to perform normalisation. Bioconductor question explaining differences between methods. Chose vst because a) it is much faster than rlog (it is recommended to use vst for samples larger than 50), and b) Michael Love (author of DESEq2) recommends using it over rlog

• Including a log fold change threshold of 0 in the results formula \(H_0:lfc=0\) because setting any other log fold change seems arbitrary and we risk losing genes with a significant differential expression for genes with a higher fold change, but not necessarily as significant.

counts = datExpr %>% as.matrix
rowRanges = GRanges(datGenes$chromosome_name,
                  IRanges(datGenes$start_position, width=datGenes$length),
                  strand=datGenes$strand,
                  feature_id=datGenes$ensembl_gene_id)
se = SummarizedExperiment(assays=SimpleList(counts=counts), rowRanges=rowRanges, colData=datMeta_sva)
dds = DESeqDataSet(se, design = ~ Batch + SV1 + SV2 + SV3 + SV4 + SV5 + SV6 + SV7 + SV8 + SV9 + 
                                  SV10 + SV11 + SV12 + SV13 + Diagnosis)

# Perform DEA
dds = DESeq(dds)

## estimating size factors

## estimating dispersions

## gene-wise dispersion estimates

## mean-dispersion relationship

## final dispersion estimates

## fitting model and testing

DE_info = results(dds, lfcThreshold=0, altHypothesis='greaterAbs')

# Perform vst
vsd = vst(dds)

datExpr_vst = assay(vsd)
datMeta_vst = colData(vsd)
datGenes_vst = rowRanges(vsd)

rm(counts, rowRanges, se, vsd)

Using the plotting function DESEq2’s manual proposes to study vst’s output it looks like the data could be homoscedastic

meanSdPlot(datExpr_vst, plot=FALSE)$gg + theme_minimal()

Plotting points individually we can notice a small heteroscedasticity in the data

plot_data = data.frame('ID'=rownames(datExpr_vst), 'Mean'=rowMeans(datExpr_vst), 'SD'=apply(datExpr_vst,1,sd))

plot_data %>% ggplot(aes(Mean, SD)) + geom_point(color='#0099cc', alpha=0.2) + geom_smooth(color = 'gray') +
              scale_x_log10() + scale_y_log10() + theme_minimal()

rm(plot_data)

Rename normalised datasets to continue working with these

datExpr = datExpr_vst
datMeta = datMeta_vst %>% data.frame
datGenes = datGenes_vst

rm(datExpr_vst, datMeta_vst, datGenes_vst, datMeta_sva)

---

Batch Effect Correction

By including the surrogate variables in the DESeq formula we only modelled the batch effects into the DEA, but we didn’t actually correct them from the data, for that we need to use ComBat (or other equivalent package) in the already normalised data

SVA surrogate variables

In some places they say you shouldn’t correct these effects on the data because you risk losing biological variation, in others they say you should because they introduce noise to the data. The only thing everyone agrees on is that you shouldn’t remove them before performing DEA but instead include them in the model.

Based on the conclusions from Practical impacts of genomic data “cleaning” on biological discovery using surrogate variable analysis it seems like it may be a good idea to remove the batch effects from the data and not only from the DE analysis:

• Using SVA, ComBat or related tools can increase the power to identify specific signals in complex genomic datasets (they found “greatly sharpened global and gene-specific differential expression across treatment groups”)

• But caution should be exercised to avoid removing biological signal of interest

• We must be precise and deliberate in the design and analysis of experiments and the resulting data, and also mindful of the limitations we impose with our own perspective

• Open data exploration is not possible after such supervised “cleaning”, because effects beyond those stipulated by the researcher may have been removed
  

Comparing data with and without surrogate variable correction

# Taken from https://www.biostars.org/p/121489/#121500
correctDatExpr = function(datExpr, mod, svs) {
  X = cbind(mod, svs)
  Hat = solve(t(X) %*% X) %*% t(X)
  beta = (Hat %*% t(datExpr))
  rm(Hat)
  gc()
  P = ncol(mod)
  return(datExpr - t(as.matrix(X[,-c(1:P)]) %*% beta[-c(1:P),]))
}

pca_samples_before = datExpr %>% t %>% prcomp
pca_genes_before = datExpr %>% prcomp

# Correct
mod = model.matrix(~ Diagnosis, colData(dds))
svs = datMeta %>% dplyr::select(SV1:SV13) %>% as.matrix
datExpr_corrected = correctDatExpr(as.matrix(datExpr), mod, svs)

pca_samples_after = datExpr_corrected %>% t %>% prcomp
pca_genes_after = datExpr_corrected %>% prcomp

rm(correctDatExpr)

Samples

Removing batch effects has a big impact in the distribution of the samples, separating them by diagnosis almost perfectly just using the first principal component

pca_samples_df = rbind(data.frame('ID'=colnames(datExpr), 'PC1'=pca_samples_before$x[,1],
                                  'PC2'=pca_samples_before$x[,2], 'corrected'=0),
                       data.frame('ID'=colnames(datExpr), 'PC1'=pca_samples_after$x[,1],
                                  'PC2'=pca_samples_after$x[,2], 'corrected'=1)) %>%
                 left_join(datMeta %>% mutate('ID'=rownames(datMeta)), by='ID')

ggplotly(pca_samples_df %>% ggplot(aes(PC1, PC2, color=Diagnosis)) + geom_point(aes(frame=corrected, id=ID), alpha=0.75) + 
         xlab(paste0('PC1 (corr=', round(cor(pca_samples_before$x[,1],pca_samples_after$x[,1]),2),
                     '). % Var explained: ', round(100*summary(pca_samples_before)$importance[2,1],1),' to ',
                     round(100*summary(pca_samples_after)$importance[2,1],1))) +
         ylab(paste0('PC2 (corr=', round(cor(pca_samples_before$x[,2],pca_samples_after$x[,2]),2),
                     '). % Var explained: ', round(100*summary(pca_samples_before)$importance[2,2],1),' to ',
                     round(100*summary(pca_samples_after)$importance[2,2],1))) +
         ggtitle('Samples') + theme_minimal())

rm(pca_samples_df)

Genes

It seems like the sva correction preserves the mean expression of the genes and erases almost everything else (although what little else remains is enough to characterise the two Diagnosis groups pretty well using only the first PC)

*Plot is done with only 10% of the genes so it’s not that heavy

pca_genes_df = rbind(data.frame('ID'=rownames(datExpr), 'PC1'=pca_genes_before$x[,1],
                                'PC2'=pca_genes_before$x[,2], 'corrected'=0, 'MeanExpr'=rowMeans(datExpr)),
                     data.frame('ID'=rownames(datExpr), 'PC1'=pca_genes_after$x[,1],
                                'PC2'=pca_genes_after$x[,2], 'corrected'=1, 'MeanExpr'=rowMeans(datExpr)))

keep_genes = rownames(datExpr) %>% sample(0.1*nrow(datExpr))

pca_genes_df = pca_genes_df %>% filter(ID %in% keep_genes)

ggplotly(pca_genes_df %>% ggplot(aes(PC1, PC2,color=MeanExpr)) + 
         geom_point(alpha=0.3, aes(frame=corrected, id=ID)) +
         xlab(paste0('PC1 (corr=', round(cor(pca_genes_before$x[,1],pca_genes_after$x[,1]),2),
                     '). % Var explained: ', round(100*summary(pca_genes_before)$importance[2,1],1),' to ',
                     round(100*summary(pca_genes_after)$importance[2,1],1))) +
         ylab(paste0('PC2 (corr=', round(cor(pca_genes_before$x[,2],pca_genes_after$x[,2]),2),
                     '). % Var explained: ', round(100*summary(pca_genes_before)$importance[2,2],1),' to ',
                     round(100*summary(pca_genes_after)$importance[2,2],1))) +
         scale_color_viridis() + ggtitle('Genes') + theme_minimal())

rm(pca_samples_before, pca_genes_before, mod, svs, pca_samples_after, pca_genes_after, pca_genes_df, keep_genes)

Everything looks good, so we’re keeping the corrected expression dataset

datExpr = datExpr_corrected

rm(datExpr_corrected)

Processing date

Even after correcting the dataset for the surrogate variables found with sva, there is still a difference in mean expression by processing date (although this difference is relatively small)

plot_data_b1 = data.frame('Mean'=colMeans(datExpr[,datMeta$RNAExtractionBatch=='10/10/2014']), 
                          'Batch'='10/10/2014')
plot_data_b2 = data.frame('Mean'=colMeans(datExpr[,datMeta$RNAExtractionBatch=='6/20/2014']), 
                          'Batch'='6/20/2014')

plot_data = rbind(plot_data_b1, plot_data_b2)
mu = plot_data %>% group_by(Batch) %>% dplyr::summarise(BatchMean=mean(Mean))

## `summarise()` ungrouping output (override with `.groups` argument)

ggplotly(plot_data %>% ggplot(aes(x=Mean, color=Batch, fill=Batch)) + geom_density(alpha=0.3) + 
         geom_vline(data=mu, aes(xintercept=BatchMean, color=Batch), linetype='dashed') +
         ggtitle('Mean expression by sample grouped by processing date') + scale_x_log10() + theme_minimal())

rm(plot_data_b1, plot_data_b2, plot_data, mu)

Performing Batch Correction for processing date

https://support.bioconductor.org/p/50983/

datExpr = datExpr %>% as.matrix %>% ComBat(batch=datMeta$Batch)

## Found2batches

## Adjusting for0covariate(s) or covariate level(s)

## Standardizing Data across genes

## Fitting L/S model and finding priors

## Finding parametric adjustments

## Adjusting the Data

Now both batches have the same mean expression

plot_data_b1 = data.frame('Mean'=colMeans(datExpr[,datMeta$RNAExtractionBatch=='10/10/2014']),
                          'Batch'='10/10/2014')
plot_data_b2 = data.frame('Mean'=colMeans(datExpr[,datMeta$RNAExtractionBatch=='6/20/2014']),
                          'Batch'='6/20/2014')

plot_data = rbind(plot_data_b1, plot_data_b2)
mu = plot_data %>% group_by(Batch) %>% dplyr::summarise(BatchMean=mean(Mean))

## `summarise()` ungrouping output (override with `.groups` argument)

ggplotly(plot_data %>% ggplot(aes(x=Mean, color=Batch, fill=Batch)) + geom_density(alpha=0.3) + 
         geom_vline(data=mu, aes(xintercept=BatchMean, color=Batch), linetype='dashed') +
         ggtitle('Mean expression by sample grouped by processing date') + scale_x_log10() + theme_minimal())

rm(plot_data_b1, plot_data_b2, plot_data, mu)

---

Save preprocessed dataset

save(datExpr, datMeta, datGenes, DE_info, dds, file='./../Data/preprocessed_data.RData')

---

Session info

sessionInfo()

## R version 3.6.3 (2020-02-29)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 18.04.4 LTS
## 
## Matrix products: default
## BLAS:   /usr/lib/x86_64-linux-gnu/blas/libblas.so.3.7.1
## LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.7.1
## 
## locale:
##  [1] LC_CTYPE=en_GB.UTF-8       LC_NUMERIC=C              
##  [3] LC_TIME=en_GB.UTF-8        LC_COLLATE=en_GB.UTF-8    
##  [5] LC_MONETARY=en_GB.UTF-8    LC_MESSAGES=en_GB.UTF-8   
##  [7] LC_PAPER=en_GB.UTF-8       LC_NAME=C                 
##  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
## [11] LC_MEASUREMENT=en_GB.UTF-8 LC_IDENTIFICATION=C       
## 
## attached base packages:
## [1] parallel  stats4    stats     graphics  grDevices utils     datasets 
## [8] methods   base     
## 
## other attached packages:
##  [1] kableExtra_1.1.0            knitr_1.28                 
##  [3] expss_0.10.2                dendextend_1.13.4          
##  [5] vsn_3.52.0                  WGCNA_1.69                 
##  [7] fastcluster_1.1.25          dynamicTreeCut_1.63-1      
##  [9] sva_3.32.1                  genefilter_1.66.0          
## [11] mgcv_1.8-31                 nlme_3.1-147               
## [13] DESeq2_1.24.0               SummarizedExperiment_1.14.1
## [15] DelayedArray_0.10.0         BiocParallel_1.18.1        
## [17] matrixStats_0.56.0          Biobase_2.44.0             
## [19] GenomicRanges_1.36.1        GenomeInfoDb_1.20.0        
## [21] IRanges_2.18.3              S4Vectors_0.22.1           
## [23] BiocGenerics_0.30.0         biomaRt_2.40.5             
## [25] ggpubr_0.2.5                magrittr_1.5               
## [27] ggExtra_0.9                 GGally_1.5.0               
## [29] gridExtra_2.3               viridis_0.5.1              
## [31] viridisLite_0.3.0           RColorBrewer_1.1-2         
## [33] plotlyutils_0.0.0.9000      plotly_4.9.2               
## [35] glue_1.4.1                  reshape2_1.4.4             
## [37] forcats_0.5.0               stringr_1.4.0              
## [39] dplyr_1.0.0                 purrr_0.3.4                
## [41] readr_1.3.1                 tidyr_1.1.0                
## [43] tibble_3.0.1                ggplot2_3.3.2              
## [45] tidyverse_1.3.0            
## 
## loaded via a namespace (and not attached):
##   [1] readxl_1.3.1           backports_1.1.8        Hmisc_4.4-0           
##   [4] plyr_1.8.6             lazyeval_0.2.2         splines_3.6.3         
##   [7] crosstalk_1.1.0.1      digest_0.6.25          foreach_1.5.0         
##  [10] htmltools_0.4.0        GO.db_3.8.2            fansi_0.4.1           
##  [13] checkmate_2.0.0        memoise_1.1.0          doParallel_1.0.15     
##  [16] cluster_2.1.0          limma_3.40.6           annotate_1.62.0       
##  [19] modelr_0.1.6           prettyunits_1.1.1      jpeg_0.1-8.1          
##  [22] colorspace_1.4-1       blob_1.2.1             rvest_0.3.5           
##  [25] haven_2.2.0            xfun_0.12              hexbin_1.28.1         
##  [28] crayon_1.3.4           RCurl_1.98-1.2         jsonlite_1.7.0        
##  [31] impute_1.58.0          iterators_1.0.12       survival_3.1-12       
##  [34] gtable_0.3.0           zlibbioc_1.30.0        XVector_0.24.0        
##  [37] webshot_0.5.2          scales_1.1.1           DBI_1.1.0             
##  [40] miniUI_0.1.1.1         Rcpp_1.0.4.6           xtable_1.8-4          
##  [43] progress_1.2.2         htmlTable_1.13.3       foreign_0.8-76        
##  [46] bit_1.1-15.2           preprocessCore_1.46.0  Formula_1.2-3         
##  [49] htmlwidgets_1.5.1      httr_1.4.1             acepack_1.4.1         
##  [52] ellipsis_0.3.1         farver_2.0.3           pkgconfig_2.0.3       
##  [55] reshape_0.8.8          XML_3.99-0.3           nnet_7.3-14           
##  [58] dbplyr_1.4.2           locfit_1.5-9.4         labeling_0.3          
##  [61] tidyselect_1.1.0       rlang_0.4.6            later_1.0.0           
##  [64] AnnotationDbi_1.46.1   munsell_0.5.0          cellranger_1.1.0      
##  [67] tools_3.6.3            cli_2.0.2              generics_0.0.2        
##  [70] RSQLite_2.2.0          broom_0.5.5            evaluate_0.14         
##  [73] fastmap_1.0.1          yaml_2.2.1             bit64_0.9-7           
##  [76] fs_1.4.0               mime_0.9               xml2_1.2.5            
##  [79] compiler_3.6.3         rstudioapi_0.11        curl_4.3              
##  [82] png_0.1-7              affyio_1.54.0          ggsignif_0.6.0        
##  [85] reprex_0.3.0           geneplotter_1.62.0     stringi_1.4.6         
##  [88] highr_0.8              lattice_0.20-41        Matrix_1.2-18         
##  [91] vctrs_0.3.1            pillar_1.4.4           lifecycle_0.2.0       
##  [94] BiocManager_1.30.10    cowplot_1.0.0          data.table_1.12.8     
##  [97] bitops_1.0-6           httpuv_1.5.2           affy_1.62.0           
## [100] R6_2.4.1               latticeExtra_0.6-29    promises_1.1.0        
## [103] codetools_0.2-16       assertthat_0.2.1       withr_2.2.0           
## [106] GenomeInfoDbData_1.2.1 hms_0.5.3              grid_3.6.3            
## [109] rpart_4.1-15           rmarkdown_2.1          shiny_1.4.0.2         
## [112] lubridate_1.7.4        base64enc_0.1-3